Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
3433632 | American Journal of Obstetrics and Gynecology | 2014 | 4 Pages |
In medicine in general, and in obstetrics in particular, it is common practice to use arbitrary cutoffs in expressing continuous variables. However, the dichotomy of continuous data is associated with loss of statistical power, which may result in inaccurate estimates in clinical prognosis or prediction of outcomes and, consequently, may lead to incorrect inferences. If the predictor is a continuous variable, arbitrary percentile-based categorizations without clinical justification is an anathema (Greek word meaning “curse”) and should be discouraged. Instead, the clinical outcome of interest should be defined first and then a receiver operating characteristic curve analysis or other appropriate statistical techniques should be employed to determine the most optimal cutoff of the predictor. The next step should be to validate the cutoff in a different population before introducing it to clinical use or interventional trials.